Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
INFM 718A / LBSC 705 Information For Decision Making Selective Perception Lecture 13 One Perspective • “We do not first see, then define, we define first and then see.” Walter Lippmann (quoted in Plous, 1993.) Selective Perception • Perception is affected by expectations. • Bruner and Postman’s (1949) experiments. • What did they find? 1 Normal Cards vs. Trick Cards Four Strategies to Cope • Dominance • Trick cards had wrong (inverse) colors, such as a black three of hearts. (Check the cover of the Plous book.) • Compromise • Disruption • Bruner and Postman found that it took people about four times longer to recognize the trick cards than normal cards. • Recognition Dominance Compromise • These people saw a red three of hearts or a black three of spades. • These people reported a red six of spades as a purple six of spades (or of hearts). • A black four of hearts was reported as a greyish four of spades. A red six of clubs was reported as “six of clubs illuminated by red light.” • 50% showed compromise behavior to red trick cards; 11% to black trick cards. • In the first case form is dominant and color is fit to expectation; in the second case vice versa. • Bruner and Postman called this “Perceptual Denial.” 2 Disruption Recognition • Some people had trouble forming any perception at all. • Some people recognized that there was a problem. • Disruption was rare but quite dramatic. • Even then, some failed to correctly identify what was wrong. Expectations… • … can strongly influence perceptions. • We probably knew that already. However, we saw that empirical evidence supports the hypothesis. How many ‘f’s in this phrase? • These functional fuses have been developed after years of scientific investigation of electric phenomena, combined with the fruit of long experience in the part of the two investigators who have come forward with them for our meetings today. • Can more experience with the subject topic strengthen the influence of expectation on perception? 3 Experience Æ Expectations Æ Perception • The second example: estimating the number of ‘f’s in a phrase. • Non-native speakers of English perform better. (Do they?) • The initial “Why?” is not completely answered. A more sophisticated experiment… • … by McMillen, Smith and Wells-Parker (1989) • “High sensation seekers” who believed they had consumed alcohol drove more recklessly than those who believed they had not. • “Low sensation seekers” who believed they had consumed alcohol drove more cautiously than those who believed they had not. Potent Expectations • Experiment by Wilson and Abrams (1977) • Heart rate was affected by whether the subject believed he was given alcohol more than whether he was actually given alcohol. • Expectations turned out to be more important than changes in blood chemistry! Dartmouth vs. Princeton (1951) • Hastorf and Cantril (1954) concluded that “It is inaccurate and misleading to say that different people have different ‘attitudes’ concerning the same ‘thing.’ For the ‘thing’ simply is not the same for different people…”1 1) Hastorf, A.H., Cantril, H., 1954, “They saw a game: A case study,” The Journal of Abnormal and Social Psychology, 49, 129-134. 4 Hostile Media Effect • Vallone, Ross and Lepper (1985) studied the 1980 U.S. presidential elections. • Approximately 1/3 of the 160 subjects (registered voters) felt that the media had been biased. In ~90% of these cases, respondents felt the bias was against the candidate they supported.2 Some Conclusions • Perceptions are selective by nature. • Perception depends on cognitive and motivational factors. • Decision makers should question their own motivations and expectations while making judgments and decisions. 2) Vallone, R.P., Ross, L., Lepper, M.R., 1985, “The hostile media phenomenon: Biased perception and perceptions of media bias in coverage of the Beirut massacre,” Journal of Personality and Social Psychology, 49, 577-585. Cognitive Dissonance • Festinger and Carlsmith’s (1959) experiment. (Tedious tasks.) Cognitive Dissonance 5 I would lie… • … for twenty dollars or for free, but not for one dollar! • So, I make myself believe that the tasks were indeed enjoyable if I am paid only $1. • Cognitive dissonance says that people try to reduce or avoid psychological inconsistencies. Self-Perception Theory • Festinger regarded cognitive dissonance as a negative motivation factor, one that should be avoided. • Bem disagreed and argued that people build their beliefs from observing themselves behave. Self-Perception Theory So what happened… • Argues that people build their beliefs and attitudes based on how they behave under various situations. • … in Festinger and Carlsmith’s experiment? • $1 case subjects looked at their own behaviors and concluded that they should have enjoyed the tasks, … since they would not lie for $1. • $20 case subjects concluded that they bent the truth a little for the money involved. • People tend to do that more when their internal cues about the situation are weak. (i.e. they do not have preconceived attitudes about the situation.) 6 The Difference • Cognitive Dissonance Theory attributes the findings to a motivation to reduce inner conflict. • Self-Perception Theory explains the findings in terms of how people infer the causes of their behaviors. Two Main Types of Dissonance • Pre-Decisional Dissonance – Sherman and Gorkin’s (1980) experiment. – Kantola, Syme, and Campbell’s (1984) study. – Doob et al.’s (1969) experiment. (Mouthwash.) • Post-Decisional Dissonance – Knox and Inkster’s (1968) survey. (Horse bets.) – Frenkel and Doob’s survey (1976) (Elections.) Conclusions • Marketing: Explicitly label introductory offers. • Politics – Social Activism: Solicit small contributions to campaigns and social causes. • Aronson’s (1972) statements. • Changes in attitude can follow changes in behavior. Memory and Hindsight Biases 7 Is Memory Reconstructive? Is Memory Reconstructive? • Myers’ (1990) example. • Loftus and Palmer’s (1974) experiments. • “Close your eyes and recall a scene in which you experienced something pleasurable.” • “How fast were the cars going?” • “Did you see any broken glass?” • …when the cars smashed, bumped, etc? Are Memories Stored Separately? Hindsight Bias • Bransford and Franks (1971) (Ants) • “I-knew-it-all-along” effect • There is an element of “memory construction” when remembering. • Elections, medical decisions, buying decisions, games, etc. • Do you have your own examples? 8 How to Reduce Hindsight Bias • Consider reasons why results might have turned out differently. Plasticity • Slovic and Fischhoff’s (1977) study. Plasticity • “Plasticity … refers to a discrepancy in how people answer two versions of the same question.”1 Example • Alternative A: Losing $50 with p=1.00. • Alternative B: Losing $200 with p=0.25 and losing nothing with p=0.75. • EV(Alternative A) = EV(Alternative B) • About 80% choose Alternative B. 1) Plous, Scott, 1993, The Psychology of Judgment and Decision Making, pp.58, McGraw-Hill, New York, NY. 9 However • When put in an insurance premium context, about 65% choose Alternative A. • Possible reasons: – Insurance premium context emphasizes the potential loss ($200), thus making it look big. – Buying insurance is considered a prudent behavior socially. – …? Question Order Effects From Schumann and Presser (1981) About half of the respondents were asked the two questions in the order below: • 1) Do you think a Communist country like Russia should let American newspaper reporters come in and send back to America the news as they see it? • 2) Do you think the United States should let Communist newspaper reporters from other countries come in and send back to their papers the news as they see it? Order Effects • The order of consecutive questions, or the order of response alternatives to questions may affect the responses by individuals. • Question order effects • Response alternative order effects Question Order Effects The other half of the respondents were asked the two questions in the reverse order as given below: • 1) Do you think the United States should let Communist newspaper reporters from other countries come in and send back to their papers the news as they see it? • 2) Do you think a Communist country like Russia should let American newspaper reporters come in and send back to America the news as they see it? 10 Results Yes • Case 1: – Q1 – Q2 82% 75% • Case 2: – Q1 – Q2 Response Alternative Order Effects • These effects are slighter than the question order effects. • A common response order effect is a type of recency effect, where respondents tend to choose the last response alternative. 55% 64% Example From Schumann and Presser (1981) • Version 1: Should divorce in this country be easier to obtain, more difficult to obtain, or stay as it is now? • Respond percentages: 23%, 36%, and 41% • Version 2: Should divorce in this country be easier to obtain, stay as it is now, or more difficult to obtain? • Respond percentages: 26%, 29%, and 46% What Can be Done? • What can be done to overcome order effects, (considering that we have to put the questions and the responses in one order or another)? In both cases, the most popular alternative was the last one. 11 Pseudo-Opinions • Some respondents tend to offer opinions even on issues about which they know very little. • Such respondents tend to shape their “pseudo-opinions” (since they do not have “real opinions”) according to how the question is asked. Example From Gill (Tide Magazine, 1947) • What is your opinion of the Metallic Metals Act? – Good move for U.S. – Should be left to individual states – O.K. for foreign states, but should not be required in U.S. – Of no value at all. • 70% of the respondents gave an opinion although there is no Metallic Metals Act! Example From Hartley (1946) • How Close Do You Fell to These Nationalities? – Danireans – Pireneans – Wallonians • More than 80% of the respondents rated these “nationalities” even though they do not exist! Pseudo-Op.s In Political Affairs • Pseudo-opinions can become important factors in political affairs. • In general, about 30% of respondents offer pseudo-opinions. • 30% can change the result of almost any major election and referendum. • Pseudo-opinions are particularly common in issues concerning foreign and military policy. 12 Filtering Pseudo-Opinions • Offering response alternatives such as “No opinion” or “I don’t know” can help filter out pseudo-opinions. • However, there may be a trap here. People may try to look like they know about the issue for social reasons, even if they do not. • “Choose not to offer opinion at this time” may be a better alternative. Saves face. Inconsistency • “Inconsistency refers to a discrepancy between two related attitudes (attitudeattitude inconsistency) or between an attitude and a corresponding behavior (attitude-behavior inconsistency).”1 1) Plous, Scott, 1993, The Psychology of Judgment and Decision Making, pp.58, McGraw-Hill, New York, NY. A-A Inconsistency Example A-B Inconsistency Example From Prothro and Grigg (1960) From Darley and Batson (1973) • Democratic principles and specific applications. • Random sample of registered voters in Ann Arbor, MI and Tallahassee, FL. • 51% “Only well-informed people should be permitted to vote.” • 79% “Only taxpayers should vote.” • Many seminary students who were about to give a speech on the parable of the Good Samaritan failed to help a person in need when they were rushed. 13 How Common is Inconsistency? • Wicker (1969) reviewed 46 studies and concluded that inconsistency is so common that it is more likely that attitudes will not be closely related to overt behaviors than they will be. • On the other hand, Ajzen and Fishbein (1977) argued that attitudes concerning particular actions and targets are generally good predictors of related behaviors. Question Wording and Framing • The way a question is worded or framed may affect individuals’ responses to that question. Wording and Framing Effects Safe and Safer • 40% of the respondents of a poll in Britain said the nuclear weapons of their country made them “feel safe.” • When the question was asked slightly differently, 50% of the respondent said the nuclear weapons of their country made them “feel safer.” 14 Another Example From Converse and Schuman (1970) • “Should the U.S. Army withdraw from Vietnam faster or slower?” Disadvantage of Middle Categories • Middle categories may act as a “safe” alternative since they are “midway” and attract pseudo-opinions. – 42% “faster,” 16% “slower,” 29% “same as now.” • “Is the pace with which the U.S. Army withdraws from Vietnam too fast, too slow, or about right?” – 29% “too slow,” 6% “too fast,” 49% “about right.” Open Ended vs. Listed Responses • Some responses may be chosen by substantially more respondents when they are listed as specific alternatives than in an open question, (a question without specific response alternatives.) Examples • Schuman and Scott’s (1987) study: “The most important problem facing this country today?” • Schwarz, Hippler, Deutsch and Strack’s (1985) study: “How much TV do you watch daily?” 15 Hilarious Examples “Marketing Application” Example From Harris (1973) • “How long was the movie?” – (Mean Answer = 130 min.) • “How short was the movie?” – (Mean Answer = 100 min.) • “How tall was the basketball player?” – (Mean Answer = 79’’) • “How short was the basketball player?” – (Mean Answer = 69’’) Social Desirability • People tend to choose responses they judge to be “socially desirable” over those they judge not to be so. Examples From Clymer (1982) • Nuclear weapons freeze From Budiansky (1988) • Support for Nicaraguan rebels From Schuman and Presser (1981) • Sending troops in case of a Vietnam-like situation. 16 Allow or Forbid Other Examples From Rugg (1941) Schuman and Presser (1981) • “Do you think that the U.S. should allow public speeches against democracy?” • Repetition of Rugg’s experiment. – 62% “No.” Hippler and Schwarz (1986) • “Do you think that the U.S. should forbid public speeches against democracy?” • Peep shows, X-rated films, salt on highways. – 46% “Yes.” • At least 16% would “not allow” rather than “forbid”! Framing When Combined • • • • Decision 1: Alternative A: Gain $240, p=1.00 Alternative B: Gain $1000, p=0.25 84% chose A. • A & D: Lose $760, p=0.75; Gain $240, p=0.25 • B & C: Lose $750, p=0.75; Gain $250, p=0.25 • • • • Decision 2: Alternative C: Lose $750, p=1.00 Alternative D: Lose $1000, p=0.75 87% chose D. • B & C are better than A & D when combined, but far more people preferred A & D over B & C. 17 Other Examples Tversky and Kahneman (1981) • Alternative programs to fight a disease. Schelling (1981) • Child tax cuts vs. “no-child tax”. Psychological Accounting • People tend to frame outcomes as well as choices. Tversky and Kahneman called this “psychological accounting.” • Example: Losing your ticket vs. losing money. (Tversky and Kahneman, 1981.) • Example: Calculator on sale vs. jacket on sale. (Tversky and Kahneman, 1981.) Conclusions • People’s answer are a function of (among other things): – Question order – Question context – Question format (open vs. closed) – Whether the pseudo-opinions were filtered – Existence of catch phrases in questions Conclusions • …: – Range of suggested response alternatives – Order of the response alternatives – Existence of middle categories as response alternatives – Framing of problems (gains vs. losses) • It would be wise not to take the findings of surveys at face value, but approach them from a critical perspective instead. 18 Expected Value Expected Utility Theory Expected Value 1: 0.166 2: 0.166 3: 0.166 Throw die Heads: 0.5 4: 0.166 5: 0.166 6: 0.166 Flip coin 1: 0.166 2: 0.166 Tails: 0.5 Throw die 3: 0.166 4: 0.166 5: 0.166 6: 0.166 • Example: Flip a coin and throw a die. If the coin shows heads and the die show an even number you get $2 times the number on the die. Otherwise you get nothing. Expected Value 0.5 * 0.166 = 0.083 0.5 * 0.166 = 0.083 0.5 * 0.166 = 0.083 0.5 * 0.166 = 0.083 • 0.083*$2*2 + 0.083*$2*4 + 0.083*$2*6 = $2 0.5 * 0.166 = 0.083 0.5 * 0.166 = 0.083 0.5 * 0.166 = 0.083 0.5 * 0.166 = 0.083 0.5 * 0.166 = 0.083 0.5 * 0.166 = 0.083 0.5 * 0.166 = 0.083 0.5 * 0.166 = 0.083 19 St. Petersburg Paradox Utility Utility • ½ * $2 + ¼ * $4 + 1/8 * 8 + … + 1/K * ($2)K = $1 + $1 + $1 + … + $1 = ∞ • How much would you pay to play this game? Wealth Expected Utility • Sum of the utilities of all possible outcomes of a chance event. Σ pi * ui Principles of Expected Utility Th. • • • • • • Ordering alternatives Dominance Cancellation Transitivity Continuity Invariance 20 Allais Paradox • Situation 1: – Alternative A: p=1, $1,000,000 – Alternative B: p=.10, $2,500,000; p=.89, $1,000,000; p=.01, $0 Paradoxes in Rationality • Situation 2: – Alternative A: p=.11, $1,000,000; p=.89, $0 – Alternative B: p=.10, $2,500,000; p=.9, $0 Allais Paradox Ellsberg’s Paradox 10 89 1 $ 1M $ 1M $ 1M 10 89 1 $ 2.5M $ 1M $0 10 89 1 $ 1M $0 $ 1M 10 89 1 $ 2.5M $0 $0 A Situation 1 30 BALLS B Betting Alternatives Red Black Yellow 1: A red ball $100 $0 $0 $0 $100 $0 2: A black ball A Situation 2 B 60 BALLS 30 BALLS 60 BALLS Betting Alternatives Red Black Yellow 1: A red or yellow ball $100 $0 $100 $0 $100 $100 2: A black or yellow ball 21 Intransitivity Intransitivity Dimensions Applicants IQ Experience A 120 1 year B 110 2 years C 100 3 years Gamble Probability of a Win A 7/24 Payoff $5.00 Expected Value $1.46 B 8/24 $4.75 $1.58 C 9/24 $4.50 $1.69 D 10/24 $4.25 $1.77 E 11/24 $4.00 $1.83 Intransitivity Preference Reversals Committee Members Candidates Ann Bob Cindy Don Ellen Joe Schmoe 1 1 2 3 3 Jane Doe 2 3 3 1 1 Al Einstein 3 2 1 2 2 Pair High Probability EV 1 .99 win $4.00 .01 Lose $1.00 .95 win $2.50 .05 Lose $0.75 … $3.95 .33 win $16.00 .67 Lose $2.00 $2.34 .40 win $8.50 .60 Lose $1.50 … … 2 … High Payoff EV $3.94 $2.50 … 22 Are Violations of EUT Irrational? • No information about the cost of errors versus the cost of following the principles. Decision Making Models • A non-logical strategy may be rational if it provides a good approximation to the utility provided by normative methods, in the long run. Satisficing • People “satisfice” rather than optimize for several reasons: Prospect Theory • Assumes that value functions for gains and losses are different. – They do not have complete information – They do not understand the available information – They cannot calculate precisely –… 23 Prospect Theory • Problem 1: – Alternative A: p=.50, gain $1000 – Alternative B: p=1.00, gain $500 (84% chose this.) Decision Weights Decision Weight • Problem 2: – Alternative A: p=.50, lose $1000 (≅70% chose this.) – Alternative B: p=1.00, lose $500 Prospect Theory • Problem 1: – Alternative A: p=1/1000, win $5000 (≅ 75% chose this.) – Alternative B: p=1.00, win $5 Probability Certainty Effect • People attribute more value (utility) to eliminating risk versus reducing risk. • Russian roulette example. • Probabilistic insurance example. • Problem 2: – Alternative A: p=1/1000, lose $5000 – Alternative B: p=1.00, lose $5 (>80% chose this.) 24 Pseudocertainty Effect • People value eliminating risk more than reducing it even when the certainty is apparent rather than real. • Vaccine example. • Dry cleaner example. Regret Theory • Based on two basic assumptions: – People feel rejoice or regret according to the outcomes of their decisions. – People try to anticipate and take account of these sensations when making decisions. • Regret theory can explain many of the paradoxes we discussed above. 25